Enhancing Target Recognition Performance in SSVEP-Based Brain-Computer Interfaces via Deep Neural Networks With...

Area of Science:

• Neuroscience and Biomedical Engineering
• Brain-Computer Interfaces (BCI)
• Signal Processing

Background:

• Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCI) offer high performance but face challenges in data scarcity and cross-subject information migration.
• Current research struggles to fully leverage existing subject data for common information mining and transfer, especially in low-data scenarios.

Purpose of the Study:

• To propose a novel deep neural network, the Pyramid Squeeze Attention Deep Neural Network (PSA-DNN), for enhancing SSVEP-BCI performance.
• To address the challenge of common information migration in low-data scenarios by developing a method for mining shared information across subjects.
• To improve SSVEP target recognition accuracy and efficiency through advanced feature extraction and a staged training approach.

Main Methods:

• Fourier transformation of band-pass filtered EEG signals to obtain frequency domain information.
• A deep neural network incorporating spatial convolution for spatial domain information extraction.
• Introduction of a pyramid attention module to enhance frequency and spatial domain information quality.
• Temporal convolution for mining time domain information from EEG signals.
• A three-stage training strategy: common information learning, personalized fine-tuning, and final classification.

Main Results:

• The proposed PSA-DNN method demonstrated favorable performance on the Benchmark and BETA datasets.
• The common information migration strategy significantly improved SSVEP target recognition, particularly in data-scarce conditions.
• The staged training approach effectively facilitated both general and personalized feature learning for enhanced BCI performance.

Conclusions:

• The PSA-DNN model offers a promising approach for enhancing SSVEP-BCI by effectively migrating common information across subjects.
• This method provides a valuable methodological reference for developing robust and efficient BCIs in real-world applications with limited data.
• The findings contribute to the theoretical understanding and practical application of cross-subject learning in SSVEP-BCI research.